Abstract Opioid overdoses have increased dramatically since 1999. In 2010, 16,651 deaths were caused by a prescription opioid overdose, representing nearly 60% of all drug overdose deaths, and exceeding overdose deaths from heroin and cocaine combined. Recent modest declines in opioid-related mortality rates have been paired with large increases in heroin abuse. In this project, we propose to study the role of prescription drug coverage expansions on the rise in opioid abuse and subsequent rise in heroin use among the population not directly impacted by the expansions. Understanding this relationship provides evidence of the magnitudes of spillovers and diversion, and the potential for policy to reduce these harms. We will use a variety of rich data sets to measure opioid distribution, opioid abuse, and mechanisms driving opioid and heroin use trends. We will study the differential effects that Medicare Part D had on geographic areas with large Medicare eligibility shares, providing policy-driven variation in prescription drug coverage unrelated to confounding factors correlated with opioid abuse. In preliminary work, we have found that Part D had a large impact on opioid abuse outcomes and that this effect persists to today, explaining a large share of geographic variation in opioid abuse. We focus on opioid abuse for the non-Medicare population given that nonmedical use is an important channel leading to adverse outcomes. We will study the mechanisms driving the past and current relationship between prescription drug coverage and opioid-related outcomes. We will also study whether increased opioid access has a gateway effect which predicts longer-term transitions to heroin, and whether this relationship was exacerbated by the introduction of abuse-deterrent OxyContin in August 2010. Our outcomes include substance abuse treatments, mortality, distribution, and nonmedical use. The data on nonmedical use include the source, such as friends or relatives. These measures are informative about diversion mechanisms. Finally, recent time series correlations suggest that the introduction of abuse-deterrent OxyContin reduced opioid abuse while possibly increasing heroin use. If this relationship is causal, we should observe that places with high prescription drug coverage rates should experience relatively larger reductions in opioid abuse upon the reformulation of OxyContin and a corresponding larger increase in heroin use. Our strategy permits us to go beyond simple time series correlations, test for causal relationships, and study whether prior access to opioids predicts transitions to heroin in this new policy environment.